Adams-Bashforth and Adams-Bashforth-Moulton Methods

The Adams-Bashforth method is a multistep method. Only the four-step explicit method is
implemented in Maple. It is not clear how the four starting values w0 , ..,w3 are obtained, but it doesn't
seem to be the Runge-Kutta method of order four as suggested by the text. However, we will compare
this method to the Runge-Kutta method of order four, since both have the same order of local truncation
error, O(h5 ). We examine the method on the IVP
v
y = y K t 2 C 1, y 0 = 0.5 on [0,2], h = 0.1.
vt
> restart:with(plots):Digits:=15;
Digits := 15
> deq:=D(y)(t)=y(t)-t^2+1;
deq := D y t = y t K t 2 C 1
> init:=y(0)=.5;
init := y 0 = 0.5
We first find the exact solution, which we rewrite as a function Y.
> soln:=dsolve({deq,init},y(t));
soln := y t = 1 C 2 t C t 2 K
> Y:=unapply(rhs(soln),t);
Y := t/1 C 2 t C t 2 K

1 t
e
2

1 t
e
2

We next apply the fourth-order Runge-Kutta method.

> rk4:=dsolve({deq, init}, y(t), type=numeric,
method=classical[rk4], start=0.0,stepsize=.1);
rk4 := proc x_classical ... end proc
Now we apply the Adams-Bashforth four step method, which is a classical method in Maple
abbreviated to adambash.
> ab:=dsolve({deq, init}, y(t), type=numeric,
method=classical[adambash], start=0.0,stepsize=.1);
ab := proc x_classical ... end proc
Now let's compare our results.
> printf("
t
y
rk4
|y-rk4|
ab
|y-ab|\n");
printf("\n");
for i from 0 to 20 do
printf("%4.1f %12.7f %12.7f %12.7f %12.7f %12.7f\n",.1*i,Y(.1*
i),eval(y(t),rk4(.1*i)),abs(Y(.1*i)-eval(y(t),rk4(.1*i))),eval(y
(t),ab(.1*i)),abs(Y(.1*i)-eval(y(t),ab(.1*i))));
od;
t
y
rk4
|y-rk4|
ab
|yab|

0.0
0.5000000
0.5000000
0.0000000
0.5000000
0.0000000
0.1
0.6574145
0.6574144
0.0000002
0.6574145
0.0000000
0.2
0.8292986
0.8292983
0.0000003
0.8292986
0.0000001
0.3
1.0150706
1.0150701
0.0000005
1.0150701
0.0000005
0.4
1.2140877
1.2140869
0.0000007
1.2140869
0.0000007
0.5
1.4256394
1.4256384
0.0000010
1.4256409
0.0000015
0.6
1.6489406
1.6489394
0.0000012
1.6489452
0.0000046
0.7
1.8831236
1.8831222
0.0000015
1.8831317
0.0000081
0.8
2.1272295
2.1272278
0.0000017
2.1272418
0.0000122
0.9
2.3801984
2.3801964
0.0000020
2.3802157
0.0000172
1.0
2.6408591
2.6408567
0.0000024
2.6408822
0.0000231
1.1
2.9079170
2.9079143
0.0000027
2.9079471
0.0000301
1.2
3.1799415
3.1799385
0.0000031
3.1799798
0.0000382
1.3
3.4553517
3.4553482
0.0000035
3.4553994
0.0000477
1.4
3.7324000
3.7323961
0.0000039
3.7324589
0.0000588
1.5
4.0091555
4.0091511
0.0000043
4.0092272
0.0000717
1.6
4.2834838
4.2834790
0.0000048
4.2835705
0.0000867
1.7
4.5530263
4.5530210
0.0000053
4.5531303
0.0001040
1.8
4.8151763
4.8151704
0.0000058
4.8153003
0.0001240
1.9
5.0670528
5.0670464
0.0000064
5.0671999
0.0001471
2.0
5.3054720
5.3054650
0.0000070
5.3056456
0.0001736
In this case, at least, it appears that the Runge-Kutta method of order 4 is superior to the AdamsBashforth method of four steps. Let's now use this method as a predictor for the three-step AdamsMoulton method to get an Adams-Bashforth-Moulton predictor-corrector method. This is also a
classical method and is abbreviated as abmoulton. The option corrections gives the number of times the
corrector is applied, the default being 1. Increasing corrections in this case does not help. In any case, it
is rarely helpful to have it greater than 4.
> abm:=dsolve({deq, init}, y(t), type=numeric,
method=classical[abmoulton], start=0.0,stepsize=.1,
corrections=1);
abm := proc x_classical ... end proc
Now let's compare our results.
> printf("
t
y
|y-ab|\n");
printf("\n");
for i from 0 to 20 do

rk4

|y-rk4|

ab

>

printf("%4.1f %12.7f %12.7f %12.7f %12.7f %12.7f\n",.1*i,Y(.1*

i),eval(y(t),rk4(.1*i)),abs(Y(.1*i)-eval(y(t),rk4(.1*i))),eval(y
(t),abm(.1*i)),abs(Y(.1*i)-eval(y(t),abm(.1*i))));
od;
t
y
rk4
|y-rk4|
ab
|yab|
0.0
0.5000000
0.5000000
0.0000000
0.5000000
0.0000000
0.1
0.6574145
0.6574144
0.0000002
0.6574145
0.0000000
0.2
0.8292986
0.8292983
0.0000003
0.8292986
0.0000001
0.3
1.0150706
1.0150701
0.0000005
1.0150701
0.0000005
0.4
1.2140877
1.2140869
0.0000007
1.2140869
0.0000007
0.5
1.4256394
1.4256384
0.0000010
1.4256385
0.0000009
0.6
1.6489406
1.6489394
0.0000012
1.6489395
0.0000011
0.7
1.8831236
1.8831222
0.0000015
1.8831223
0.0000013
0.8
2.1272295
2.1272278
0.0000017
2.1272280
0.0000016
0.9
2.3801984
2.3801964
0.0000020
2.3801966
0.0000019
1.0
2.6408591
2.6408567
0.0000024
2.6408569
0.0000022
1.1
2.9079170
2.9079143
0.0000027
2.9079144
0.0000026
1.2
3.1799415
3.1799385
0.0000031
3.1799384
0.0000031
1.3
3.4553517
3.4553482
0.0000035
3.4553480
0.0000036
1.4
3.7324000
3.7323961
0.0000039
3.7323958
0.0000042
1.5
4.0091555
4.0091511
0.0000043
4.0091505
0.0000050
1.6
4.2834838
4.2834790
0.0000048
4.2834780
0.0000058
1.7
4.5530263
4.5530210
0.0000053
4.5530196
0.0000067
1.8
4.8151763
4.8151704
0.0000058
4.8151685
0.0000078
1.9
5.0670528
5.0670464
0.0000064
5.0670438
0.0000090
2.0
5.3054720
5.3054650
0.0000070
5.3054616
0.0000104
We see a significant improvement from the corrector, but the Runge-Kutta method of order 4 is again
superior even here. Of course, all have the same order of truncation error, so this should not be too
surprising.

NumericalAnalysis
> with(Student[NumericalAnalysis]);
AbsoluteError, AdamsBashforth, AdamsBashforthMoulton, AdamsMoulton, AdaptiveQuadrature,
AddPoint, ApproximateExactUpperBound, ApproximateValue, BackSubstitution, BasisFunctions,
Bisection, CubicSpline, DataPoints, Distance, DividedDifferenceTable, Draw, Euler, EulerTutor,
ExactValue, FalsePosition, FixedPointIteration, ForwardSubstitution, Function,

InitialValueProblem, InitialValueProblemTutor, Interpolant, InterpolantRemainderTerm,

IsConvergent, IsMatrixShape, IterativeApproximate, IterativeFormula, IterativeFormulaTutor,
LeadingPrincipalSubmatrix, LinearSolve, LinearSystem, MatrixConvergence,
MatrixDecomposition, MatrixDecompositionTutor, ModifiedNewton, NevilleTable, Newton,
NumberOfSignificantDigits, PolynomialInterpolation, Quadrature, RateOfConvergence,
RelativeError, RemainderTerm, Roots, RungeKutta, Secant, SpectralRadius, Steffensen, Taylor,
TaylorPolynomial, UpperBoundOfRemainderTerm, VectorLimit
We will solve the same problem with which we began the worksheet by using the AdamsBashforth
command, which is short for InitialValueProblem with method=adamsbashforth. We enter the IVP
after resetting some variables that have been given values.
> y=`y`;t=`t`;
y=y
t=t
> deq:=diff(y(t),t)=y(t)-t^2+1;
d
deq :=
y t = y t K t2 C 1
dt
> init:=y(0)=0.5;
init := y 0 = 0.5
This method has several submethods: step2 (local truncation error of O(h3 ), step3 (O(h4 )), step4 (O(h5 )
), and step5 (O(h6 )). We look at step4 with a step-size of 0.1 after first resetting the interface.
> interface(rtablesize=100);
10
> AdamsBashforth(deq,init,t=2,submethod=step4,numsteps=20,digits=10,
output=information,comparewith=[[rungekutta,rk4]]);

A-B 4-Step

Error

R-K 4th Ord.

Error

0.5

0.

0.5

0.

0.1000000000 0.6574145410 0.6574143750

1.660 10-7

0.6574143750

1.660 10-7

0.2000000000 0.8292986209 0.8292982760

3.449 10-7

0.8292982760

3.449 10-7

0.3000000000 1.015070596

1.015070058

5.38 10-7

1.015070058

5.38 10-7

0.4000000000 1.214087651

1.214089170 0.000001519

1.214086906

7.45 10-7

0.5000000000 1.425639365

1.425643661 0.000004296

1.425638396

9.69 10-7

0.6000000000 1.648940600

1.648948027 0.000007427

1.648939390

0.000001210

0.7000000000 1.883123646

1.883134837 0.000011191

1.883122179

0.000001467

0.8000000000 2.127229536

2.127245234 0.000015698

2.127227791

0.000001745

0.9000000000 2.380198444

2.380219470 0.000021026

2.380196402

0.000002042

1.000000000

2.640859086

2.640886381 0.000027295

2.640856724

0.000002362

1.100000000

2.907916988

2.907951641 0.000034653

2.907914285

0.000002703

1.200000000

3.179941539

3.179984795 0.000043256

3.179938470

0.000003069

1.300000000

3.455351666

3.455404953 0.000053287

3.455348207

0.000003459

1.400000000

3.732400017

3.732464964 0.000064947

3.732396141

0.000003876

1.500000000

4.009155465

4.009233937 0.000078472

4.009151145

0.000004320

1.600000000

4.283483788

4.283577911 0.000094123

4.283478996

0.000004792

1.700000000

4.553026304

4.553138503 0.000112199

4.553021009

0.000005295

1.800000000

4.815176268

4.815309302 0.000133034

4.815170440

0.000005828

1.900000000

5.067052779

5.067209791 0.000157012

5.067046386

0.000006393

2.000000000

5.305471951

5.305656512 0.000184561

5.305464960

0.000006991

0.

y t
0.5

Next we solve the problem by using the AdamsBashforth command, which is short for
InitialValueProblem with method=adamsmoulton. We enter the IVP after resetting some variables that
have been given values.
> AdamsMoulton(deq,init,t=2,submethod=step3,numsteps=20,digits=10,
output=information,comparewith=[[rungekutta,rk4]]);

A-M 3-Step

Error

R-K 4th Ord.

Error

0.5

0.

0.5

0.

0.1000000000 0.6574145410 0.6574143750

1.660 10-7

0.6574143750

1.660 10-7

0.2000000000 0.8292986209 0.8292982760

3.449 10-7

0.8292982760

3.449 10-7

0.3000000000 1.015070596

1.015070050

5.46 10-7

1.015070058

5.38 10-7

0.4000000000 1.214087651

1.214086866

7.85 10-7

1.214086906

7.45 10-7

0.5000000000 1.425639365

1.425638296 0.000001069

1.425638396

9.69 10-7

0.6000000000 1.648940600

1.648939196 0.000001404

1.648939390

0.000001210

0.7000000000 1.883123646

1.883121849 0.000001797

1.883122179

0.000001467

0.8000000000 2.127229536

2.127227278 0.000002258

2.127227791

0.000001745

0.9000000000 2.380198444

2.380195649 0.000002795

2.380196402

0.000002042

1.000000000

2.640859086

2.640855665 0.000003421

2.640856724

0.000002362

1.100000000

2.907916988

2.907912841 0.000004147

2.907914285

0.000002703

1.200000000

3.179941539

3.179936550 0.000004989

3.179938470

0.000003069

1.300000000

3.455351666

3.455345706 0.000005960

3.455348207

0.000003459

1.400000000

3.732400017

3.732392934 0.000007083

3.732396141

0.000003876

1.500000000

4.009155465

4.009147091 0.000008374

4.009151145

0.000004320

1.600000000

4.283483788

4.283473929 0.000009859

4.283478996

0.000004792

1.700000000

4.553026304

4.553014741 0.000011563

4.553021009

0.000005295

1.800000000

4.815176268

4.815162750 0.000013518

4.815170440

0.000005828

1.900000000

5.067052779

5.067037024 0.000015755

5.067046386

0.000006393

2.000000000

5.305471951

5.305453637 0.000018314

5.305464960

0.000006991

0.

y t
0.5

Again, rk4 is slightly more accurate than Adams Moulton. Finally, we use the method with a fourthorder Adams Bashforth Predictor and a fourth-order Adams Moulton corrector. One can use step 2,3,4,
or 5 predictor correctors, The default is 4.
> AdamsBashforthMoulton(deq,init,t=2,submethod=step4,numsteps=20,
digits=10,output=information,comparewith=[[rungekutta,rk4]]);

4th-Ord. A-B-M

Error

R-K 4th Ord.

Error

0.5

0.

0.5

0.

0.1000000000 0.6574145410

0.6574143750

1.660 10-7

0.6574143750

1.660 10-7

0.2000000000 0.8292986209

0.8292982760

3.449 10-7

0.8292982760

3.449 10-7

0.3000000000 1.015070596

1.015070058

5.38 10-7

1.015070058

5.38 10-7

0.4000000000 1.214087651

1.214086961

6.90 10-7

1.214086906

7.45 10-7

0.5000000000 1.425639365

1.425638498

8.67 10-7

1.425638396

9.69 10-7

0.6000000000 1.648940600

1.648939526

0.000001074

1.648939390

0.000001210

0.7000000000 1.883123646

1.883122331

0.000001315

1.883122179

0.000001467

0.8000000000 2.127229536

2.127227941

0.000001595

2.127227791

0.000001745

0.9000000000 2.380198444

2.380196525

0.000001919

2.380196402

0.000002042

1.000000000

2.640859086

2.640856792

0.000002294

2.640856724

0.000002362

1.100000000

2.907916988

2.907914262

0.000002726

2.907914285

0.000002703

1.200000000

3.179941539

3.179938314

0.000003225

3.179938470

0.000003069

1.300000000

3.455351666

3.455347869

0.000003797

3.455348207

0.000003459

1.400000000

3.732400017

3.732395563

0.000004454

3.732396141

0.000003876

1.500000000

4.009155465

4.009150258

0.000005207

4.009151145

0.000004320

1.600000000

4.283483788

4.283477718

0.000006070

4.283478996

0.000004792

1.700000000

4.553026304

4.553019248

0.000007056

4.553021009

0.000005295

1.800000000

4.815176268

4.815168084

0.000008184

4.815170440

0.000005828

1.900000000

5.067052779

5.067043310

0.000009469

5.067046386

0.000006393

2.000000000

5.305471951

5.305461016

0.000010935

5.305464960

0.000006991

0.

y t
0.5

We actually lose a touch of accuracy here.